Feb 2025 – Present · 1 yr 4 mos · Noida, Uttar Pradesh, India · On-site

• Praro – AI-Based Learning Platform: Ingested CCSS (via API) and CBSE (via LLM-parsed PDFs) curriculum data into Firestore to power intelligent autocomplete. Built a prompt evaluation pipeline with Claude 4 Sonnet, and generated adaptive quizzes via GPT-4o with WebSocket streaming for real-time UI. Added resume-from-last-question support and built a teacher-feedback loop using vector DB, improving evaluation accuracy by 70%. Designed an admin dashboard with SVG-based diagram generation. Tech Used: GPT-4o, Claude 4 Sonnet, FastAPI, Firebase, WebSocket, LangChain, Faiss DB, ReactJS.
• FlyCom Chatbot – AI-Powered SQL Assistant: Built a natural language to SQL chatbot for FlyDocs, automating query generation with GPT-4o and DDL-aware prompt engineering. Implemented fuzzy column matching, mapped integer codes to string values, and enabled multi-column, count-based, and date/time queries with CSV export. Used async FastAPI APIs and SSE for real-time responses, significantly improving UX. Tech Used: FastAPI, GPT-4o, LangChain, SQLAlchemy, FuzzyWuzzy, HTML/CSS/JS.

Dec 2023 – Feb 2025 · 1 yr 2 mos · Noida, Uttar Pradesh, India · On-site

• Virtual Consultation – DigiLawyer (Legal AI Chatbot): Built an AI-driven chatbot providing real-time legal advice across domains, with async SSE pipelines streaming 4 parallel LangChain chains (response, edge, strategy, outcomes). Designed facts extraction logic from form inputs and conversational context, and delivered structured multi-tab responses that reduced consultation time. Tech Used: GPT-4o, LangGraph, FastAPI, SSE, AsyncIO, SQLAlchemy, PostgreSQL.
• Current Status Auto Attachment (Aviation): Automated aviation document bunching, reducing manual effort by 40% using SpaCy keyword extraction and ML-based page classification (Random Forest). Built a scalable multiprocessing system in FastAPI, implemented retraining/versioning pipelines, and deployed as a Dockerized microservice with MySQL. Tech Used: FastAPI, Docker, SpaCy, scikit-learn, Azure Service Bus, MySQL, Pandas, NLTK.
• FlyTag – AI-Powered Text Classification: Developed an in-house DL-based multi-class text classification system, reducing costs by 30%. Built a FastAPI pipeline with text cleaning, tokenization, sequence padding, and probability-based tag prediction. Implemented model versioning, rollback support, dataset balancing checks, and automated evaluation reporting. Tech Used: FastAPI, TensorFlow, scikit-learn, Pandas, NLTK, Multithreading.

Aug 2023 – Dec 2023 · 4 mos · Noida, Uttar Pradesh, India · On-site

• Learnt FastAPI and implemented MVC pattern for diverse projects.
• Developed an OCR project for document recognition and information extraction.
• Conducted document verification prior to information extraction, ensuring the validity of documents such as Aadhaar, PAN, GST, Passport, Visa, etc.
• Utilized object detection techniques of computer vision with YOLO v8 model for doc image verification.
• Carefully fine-tuned a pretrained model, evaluating its performance on augmented doc image data.
• Implemented Google Cloud Vision SDK to extract raw text from document images.
• Employed regex parsing to extract key information from raw text, providing a refined response.